# GLM-5.3 (Z.ai)

LLM, Z.ai, reasoning low. Verbalized probabilities through the Jevals adapter, via OpenRouter · Novita (fp8).

| Board | Decision Score (95% range) | Rank | Accuracy | Calibration gap (ECE, pts) | Hand-off at 95% | $ per 1k decisions | Speed p95 |
| --- | --- | --- | --- | --- | --- | --- | --- |
| [noul (PubMedQA)](https://jevals.com/noul/) | 60.6 (50.0 to 69.8) | tied for rank 3 of 8 | 88.7% | 2.0 | 45% | $0.51 | 2.9 s |
| [choice (Banking77)](https://jevals.com/choice/) | 66.8 (62.0 to 71.7) | tied for rank 2 of 8 | 78.9% | 6.2 | 38% | $0.72 | 3.2 s |
| [score (HelpSteer2 helpfulness)](https://jevals.com/score/) | 7.8 (−4.6 to 18.6) | tied for rank 1 of 8 | 43.0% | 12.9 | — | $0.78 | 3.6 s |

Decision Score: 100 = perfect, 0 = guessing the label base rates, below 0 = worse than that. Ranks: 1 + the number of rows significantly better (paired item bootstrap, 95%); rows that cannot be told apart share a rank. Hand-off at 95%: the share of decisions a model can take alone, at its own confidence threshold, while staying at least 95% right.

Setup: model id z-ai/glm-5.3; served as z-ai/glm-5.3; adapter jevals-llm 0383a0e3e592; reasoning low (mandatory); temperature provider default; price as of 2026-09-18.

Compare: [Jev vs GLM-5.3](https://jevals.com/compare/jev-vs-glm-5.3/)

Source: https://jevals.com/models/glm-5.3/ · release 2026-09-18, suite 0.1.0 · data (CC-BY-4.0): https://jevals.com/data/releases/2026-09-18/board.json · cite as: Jevals (jevals.com), release 2026-09-18, suite 0.1.0. CC-BY-4.0.